Feature #1484
Fires on “ I” (not labeled yet: read the examples)
- Fires on
- 0.16% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 86%
- agent 8%
- chat 4%
- python 1%
- worksheet 0%
Makes the model say
to back not't should away just mean
Makes it not say
⏎⏎>>><|assistant_end|>lowt calculatorONor
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 1.94She felt sorry. She said, "I am sorry, daddy. I did not mean to
-
story 1.85⏎"I'm sorry, mom. I'm sorry, Lily. I was playing with my
-
story 1.69Tom feels bad. He says, "I'm sorry, Mom. I did not know.
-
story 1.67is funny.⏎"Sorry, Tim. I was joking. That is not the moon
-
story 1.66his friends did not like his new hat. Zip learned that it is not always good to
-
story 1.65are you crying?" Lily says, "I am sorry, Ben. I was fearful.
-
story 1.64wanted to make her happy. He said, "Sorry, mom. I was just bored.
-
story 1.63pages with dirty hands." Tom said, "Sorry, Lucy. I will be thoughtful
-
story 1.59shiny thing and tried to bite it.⏎"Hey, cat! That is my trump
-
story 1.56and sorry for his sister.⏎"I'm sorry, Lily. I tried to tell you
-
story 1.55. She was not happy.⏎"I'm sorry, mom. I wanted to touch the
-
story 1.55I'm sorry. I didn't know it was her baby doll. I thought it was a